Unanswered

Dynamically Filter knowledge sources(dataverse table) in Copilot Studio

(1) Share

Report

Posted on by JasonAbba

Hello Community,

I am trying filter the dataverse table based on user query.

For example: If they user asks "what is the total revenue generated previous month"

I want the copilot to filter the dataverse table used as knowledge source by month then calculate the total revenue by totalling all the values in Amount column.

If I write a query like "Please total all the values in Amount Column in Sales_Data table where the value Month column is AUG-24" it will give me a response. But as a developer I do not expect the user to know all the column names in dataverse table which will help him to write a clear and concise prompt.

The above mentioned was the first approach which is not feasible, So the second approach was to create topics

I created a topic called as Total Revenue which would trigger whenever the user typed ''total revenue''. But this approach is not open-end because what if the typed "What is the total revenue past month". Since the question contains the phrase total revenue, the particular topic would trigger giving the answer for total revenue but not for past month.

Please let me know if there a possibility of dynamically filtering the dataverse table used as knowledge source.

thanks in advance!

Categories:

General topics

I have the same question (0)

All responses (3)

Answers (0)

fchopo 8,003 Moderator on at

Like (1)

Report

Hi @JasonAbba

That's an interesting question, and it's basically related to topic design. I would recommend you to have a look at the Bot Building Handbook which contains some really useful resources about topic design.

In your specific case, the only thing that comes to my mind is that you trigger a topic when user types "total revenue", and after that, you need to ask the user the period (last month, last 3 months, current, etc), so you do the correct calculation. Not the ideal situation, but that could solve your problem.

Hope that helps!

Ferran

Was this reply helpful? Yes No
JasonAbba 51 on at

Like (0)

Report

Hello @fchopo

Thank you for sharing the resource link, I will surely go through it and come back if if I find any answers!

Regards
Jason

Was this reply helpful? Yes No
Suggested answer

sandeepstw 312 Moderator on at

Like (1)

Report

Hi,

For this, we'll need a custom architecture. If it's database-related, you'll need to handle custom query creation based on natural language inputs, using dynamically generated queries.

This can be achieved with a combination of vector databases and LLMs like Azure OpenAI and Cosmos DB.

Feel free to DM me if you'd like to discuss further.

Thanks,
Sandeep

Was this reply helpful? Yes No